Anda di halaman 1dari 10

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2606404, IEEE Access

QoE-based Distributed Multichannel Allocation in


5G Heterogeneous Cellular Networks: A
Matching-coalitional Game Solution
Ducheng Wu, Qihui Wu, Senior Member, IEEE, Yuhua Xu, Member, IEEE, Jianjun Jing, Zhiqiang Qin

Abstract—We investigate the problem of multichannel allo- capacity to meet the rapid growth of wireless data services
cation for small-cell users (SUs) in 5G heterogeneous cellular and requirements, i.e., 60 Mbps and 300 Mbps in uplink and
networks (HCNs) by taking users’ quality of experience (QoE) downlink, respectively, in 95 percent of locations and time [1].
into account. In most existing channel allocation approaches, they
are assumed that each user can only transmit on one channel Small-cell technologies are seen as a technology component
and the optimization goal is the network throughput without for the 5G wireless communication system to increase the
considering users’ QoE demands. Moreover, the individual QoE wireless data by reusing spectrum resources [2]–[5]. Small-
losses of macro-cell users (MUs), which are caused by the cross- cell base stations (SBSs) are low-power and short-range base
tiered interference, are not considered. stations providing wireless services to small-cell users (SUs)
In this paper, considering SUs’ QoE demands and the in-
dividual QoE losses of MUs, we propose a joint matching- which are different from traditional macro-cell base stations
coalitional game theoretical scheme to solve such a QoE based (MBSs). Because SBSs can be installed in surrounding ar-
multichannel allocation problem with individual cross-tiered eas, the 5G cellular wireless systems will have a multi-tier
interference constraint in each channel. Concretely, according to architecture consisting of macro-cells and small-cells to serve
the different interference and competition relationships among users. In underlay heterogeneous wireless networks, macro-
users, we divide the complicated problem into two subproblems,
i.e., Q1: intra-cell channel allocation for SUs, and Q2: inter- cells share the spectrum resource with small-cells, and macro-
cell channel allocation for small-cell base stations (SBSs). We cell users (MUs) are interfered by SBSs at the same time with
formulate the intra-cell channel allocation as a many-to-one the increasing of spectral efficiency.
selfish matching game and formulate the inter-cell channel allo- It is well known that the quality of experience (QoE) is
cation subproblem as an altruistical coalitional game separately. an essential indicator in 5G wireless communication system.
Then, the complicated problem can be solved based on the
two proposed games iteratively. We propose a joint channel In order to ensure a high degree of QoE for each MU, the
allocation algorithm for the matching-coalitional game theoretical cross-tiered interference caused by SBSs in each channel must
scheme. We prove that the proposed algorithm converges to a be limited. However, in order to increase the QoE of SUs,
stable channel allocation profile. Simulation results show that the each SBS is willing to occupy spectrum resource as much as
proposed algorithm achieves higher global SUs’ satisfaction than possible. The co-tiered interference among SBSs and the cross-
the smallest interference channel selection and random allocation.
tiered interference between MBSs and SBSs will be more
Index Terms—5G, heterogeneous cellular networks, multichan- serious. Thus, there is a lot of necessary to manage co-tiered
nel allocation, QoE, game. and cross-tiered interference to exploit the benefits promised
by SBSs [6]–[8].
I. I NTRODUCTION In this paper, we focus on the mitigation of co-tiered
interference and study the issue of QoE-based multichannel
HE traffic load in current wireless networks has exponen-
T tially increased due to the proliferation of new wireless
technologies, and will continue to increase in the foresee-
allocation for SUs with cross-tiered interference constraint for
MUs in heterogeneous wireless networks.
There are some existing works which have studied re-
able future. The next-generation (5G) wireless communication source allocation problems in 5G heterogeneous networks
system is expected to provide high data rates and system with convex optimization algorithm [10], greedy algorithm
[9], clustering based approach [11], [12] and game based
This work was supported by the Natural Science Foundation for Distin-
guished Young Scholars of Jiangsu Province under Grant No. BK20160034, approach [13]–[15]. Among them, game based approach has
the National Natural Science Foundation of China under Grant No. 61401508 been seen as one of the most important decision theoretic
and No. 61671473, and in part by the Open Research Foundation of National solutions for resource allocation in 5G communication system
Key Laboratory of Science and Technology on Information Transmission and
Dissemination Technologies in Communication Networks. due to the distributed and autonomous decision process. In
Ducheng Wu, Yuhua Xu and Jianjun Jing are with the College of [18], the authors proposed a local altruistic potential game for
Communications Engineering, PLA University of Science and Technology, distributed channel selection, and used the spatial adaptive play
Nanjing, China (e-mail: wuducheng@foxmail.com, yuhuaenator@gmail.com,
jianjun.jing@panda.cn). to achieve global optimization. In [19], the authors proposed
Qihui Wu is with the College of Electronic and Information Engineering, a MAC-layer interference minimization potential game to
Nanjing University of Aeronautics and Astronautics, Nanjing, China (e-mail: investigate the problem of distributed channel allocation. In
wuqihui2014@sina.com).
Zhiqiang Qin is with Information Science and Technology Institute, the proposed game, there is no information exchange among
Zhengzhou, China (e-mail: qinzhiq2006@163.com). users. Users in the proposed game select only one channel

2169-3536 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2606404, IEEE Access

to transmit information. In [13], [16], [17], the problem of proposed games iteratively. ii) The cross-tiered interference
resource allocation in the small-cell networks was formulated constraint can only be considered in the inter-cell channel
as a Stackelberg game, where the MBS acted as a leader allocation subproblem. The reason is that for MUs the cross-
and could put forward its own strategies and then the SBSs tiered interference is caused by SBSs. The cross-tiered inter-
acted as followers and respond. In [20], the authors proposed a ference constraint is more properly handed in the coalitional
coalition formation game based scheme to address the uplink game in which SBSs are the players. iii) Compared with
resource allocation problem for multiple D2D and cellular the user terminal equipments, SBSs have strong abilities of
users which combed different transmission modes, mutual communication and computing. In many situations, SBSs are
interferences, and resource sharing policy in a single utility deployed by communication operators. Thus, SBSs can solve
function. Literature [21] constituted a unified treatment of the inter-cell channel allocation subproblem cooperatively and
matching theory tailored to the peculiar properties of resource altruistically with the coalitional game. While, SUs with
management in 5G wireless networks. weaker communication and computing abilities are rational
However, in most of the above mentioned works, there and naturally selfish, so they want to optimize their own QoE
are three main drawbacks: i) Most mentioned works assumed satisfaction. Thus, SUs selfishly compete for the channels in
that one SU (or MU) can only occupy one channel (or a the proposed intra-cell channel allocation matching game. The
resource block) at a certain time. In practice, the demands of proposed joint matching-coalitional game not only focuses on
SUs are different and each SU can transmit in multichannel optimizing a global function at the SBS (macroeconomic) level
simultaneously due to the progress of the hardware devices. ii) but also considers SUs’ local decisions and interactions at the
In most existing works, the optimization goals were the system SU (microeconomic) level.
throughput and did not consider SUs’ QoE requirements. It is The main contributions of this article are summarized as
known that the QoE utility is a nonlinear function of the sum follows:
throughput of all occupied channels for each SU. Thus, the • Considering the individual interference to each MU, we
influence to users’ QoE, which is caused by cross-tiered and formulate the QoE based multichannel allocation problem
co-tiered interference, is more difficult to detect and many of of SUs in 5G heterogeneous underlaying cellular net-
the existing methods are no longer applicable. iii) Among the works as a nonlinear optimization problem with cross-
above mentioned works, [18], [19] only studied the problem tiered interference constraint. To simplify the compli-
of co-tiered interference mitigation in monolayer small-cell cated combinatorial optimization problem, we divide the
networks and did not consider cross-tiered interference in problem into two subproblems, i.e., i) intra-cell channel
heterogeneous networks. In [13], [16], [17], [20], cross-tiered allocation for SUs and ii) inter-cell channel allocation for
interference was modeled into different utility functions of SBSs, which have considered the different interference
SUs, SBSs and MBSs without considering the individual and competition relationships among SUs and MUs and
influence to the satisfaction of each MU. captured the different characteristics of SUs and SBSs.
In this paper, we propose a joint matching-coalitional game • To analyze and solve the two subproblems separately, we
theoretical scheme to allocate channels for SBSs and their propose two resource allocation games, i.e., many-to-one
associated SUs in heterogeneous wireless networks, which matching game and coalitional game. Combining these
makes an effort to promote the global SUs’ satisfaction with two games, we propose a joint matching-coalitional game
individual cross-tiered interference constraint to protect MUs’ theoretical scheme to solve the complicated combinatorial
QoE satisfaction. It is well known that there are three kinds optimization problem in an iterative manner.
of influence relationships among users in the heterogeneous • We propose a joint channel allocation algorithm for
wireless network, i.e., the cross-tiered interference relationship the matching-coalitional game theoretical scheme. We
between SUs and MUs, the co-tiered interference relationship prove that the proposed algorithm converges to a stable
among SUs associated with different SBSs (inter-cell SUs), channel allocation profile for both the proposed many-
and the competition relationship among SUs associated with to-one matching game and coalitional game. We evaluate
same SBSs (intra-cell SUs). Based on the different interference the influence of different network environments on our
and competition relationships, we divide the complicated chan- proposed matching-coalitional game based approach by
nel allocation problem into two subproblems, i.e., Q1: intra- comparing it with some other existing approaches.
cell channel allocation for SUs, and Q2: inter-cell channel allo- The rest of this article is organized as follows. In section
cation for SBSs. We formulate the intra-cell channel allocation II, we present the system model and problem formulation. In
as a many-to-one matching game (in which SUs selfishly Section III, the multichannel allocation problem is formulated
compete for the channels which have been allocated to SBSs) as a joint matching-coalitional game theoretical scheme, the
and formulate the inter-cell channel allocation subproblem existence of stable channel allocation profile is analyzed, and
as a coalitional game (in which SBSs utilize the available a multi-agent learning algorithm is proposed. In Section IV,
channels cooperatively and altruistically under the cross-tiered simulation results and discussion are given. Finally, we make
interference constraint) separately. conclusion in Section V.
The reasons for using the proposed joint matching-
coalitional game theoretical scheme model are threefold: i) II. S YSTEM M ODEL AND P ROBLEM F ORMULATION
The complicated channel allocation problem is divided into The model of a single macro-cell with M MUs, i.e., M =
two tractable subproblems and can be solved based on the two {1, 2, . . . , M }, and N small-cells, i.e., N = {1, 2, . . . , N }, are

2169-3536 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2606404, IEEE Access

where p0 is the transmission power of the MBS in each


l
channel, g0,m is the channel gain from the MBS to MU m
in channel l, Fl is the SBS set who have occupied channel l,
i.e., ∀n ∈ Fl , there has l ∈ Cn . pn is the transmission power
l
of SBS n, gn,m is the channel gain from SBS n to MU m in
2
channel l, σm is the noise power.
The achieved throughput of MU m can be described as:

Rm = B0 log2 (1 + γm,l ), (2)
l∈bm

where B0 is the bandwidth of each channel.


SU MU SBS signal MBS signal In current and future wireless systems, a key user-related
Cross-tiered interference Co-tiered interference metric is Quality of Experience (QoE). In the following,
we use a specific unified demand-aware satisfaction utility
Fig. 1. System model of a macro-cell and small-cells with downlink resource function to describe QoE for different users with different
sharing.
throughput requirements. The satisfaction function of users is a
universal sigmoid function, which can unify the utility function
considered as shown in Fig. 1. There are L dedicated channels, of users with different traffics [23]. And it is expressed as
i.e., L = {1, . . . , L}, for MUs to transmit information. MUs follow:
transmit information on orthogonal channels to avoid conflict. 1
u(r) = , (3)
It is assumed that small-cells can transmit in same channels 1 + e−τ (r−rreq )
with MUs only when the satisfactions of all MUs are higher where r is the received throughput of each user, τ is used
than determinate levels. On other words, the received cross- to adjust the slope of the utility curve around the required
tiered interference of MUs in their occupied channels is transmission rate rreq . τ reflects the demand degree of rreq .
smaller than some thresholds. The parameters τ and rreq are different for different users.
In each small-cell, there are one or more SUs associating This method is also applicable to other satisfaction function
with the SBS. We assume that there are S SUs in the network, forms.
i.e., S = {1, 2, . . . , S}. And πn SUs are associated with each When small-cells share the dedicated channels with MUs,
SBS n in∪a closed style, i.e., Sn = {xn1 , xn2 , . . . , xnπn }, the satisfaction of MU m (uM U
m (Rm )) must higher than a
xni ∈ S, n∈N Sn = S and Sn ∩ Sn′ = ø, ∀n ̸= n′ . It is as- predetermined value Vm = uM m
U
(R Vm ), where Vm is the QoE
sumed that SBSs can transmit in same channels with inter-cell threshold of MU m. Since u(r) is a nondecreasing function
interference, but SUs associated with the same SBS can only of r, there must be Rm > RVm , where RVm is the throughput
occupy orthogonal channels to avoid intra-cell interference. threshold of MU m. Thus, the received interference of MU
Since the allocated channel resource of MUs can be m in its occupied channels must be restrained. In order to
managed efficiently with the resource scheduling and power facilitate analysis, we assume that the received interference of
control mechanisms [22] and the macro-cell has a higher MU m caused by SBSs in all its occupied channels is smaller
priority to allocate channels to MUs, we focus on the channel than a same interference threshold Ithed,m ,
allocation problem between small-cells and SUs with cross-
p0 g0,m RVm
tiered interference constraint in the scenario that MUs have B0 log2 (1 + )= . (4)
2
Ithed,m + σm µm
been allocated determinate channels to transmit.
We set bm = {βm 1 2
, βm µm
, . . . , βm } and as = In other words, the interference thresholds of L dedicated
{αs , αs , . . . , αs } as allocated channel set of MU m
1 2 ηn
channels can be calculated before the channel allocation
and SU s, where βm j
, αsi ∈ L, 0 ≤ µm ≤ L, 0 ≤ ηn ≤ L. of small-cells. The interference thresholds are expressed as
Particularly, it is assumed that bm ∩ bm′ = ø, ∀m, m′ ∈ M. {I 1 , I 2 , . . . , I L }.
Similarly, SUs in certain small-cell n transmit information Similarly, the SINR of SU s in channel l can be described
on orthogonal channels, i.e., as ∩ as′ = ø, ∀s, s′ ∈ Sn . as
Here b = (b1 , b2 , . . . , bM ) has been determined by the pφs gφl s ,s
γs,l = l +
∑ l
, (5)
MBS, while a = (a1 , a2 , . . . , aS ) is the optimized variable p0 g0,s pn gn,s + σs2
set to be determined. Specifically, the channel selection n∈Fl \φs

strategy of SUs in certain small-cell n can be denoted as where φs is the associated SBS of SU s.
an = (axn1 , axn2 , . . . , axnπn ), an ⊂ a. It can easily find The aggregate throughput of SU s in small-cell φs can be
that the occupied channel set of SBS n can be denoted as described as:
π∪n ∑
Cn = axni , where Cn = {c1 , . . . , c|Cn | } , Cn ⊂ L. Rs = B0 log2 (1 + γs,l ). (6)
i=1 l∈as
Then, the power of received signal to interference and noise
ratio (SINR) of MU m in channel l can be described as Then the satisfaction of SU s is us (Rs ). The global QoE utility
l of SUs is given by
p0 g0,m ∑
γm,l = ∑ l 2
, (1) U (a) = us (Rs ). (7)
pn gn,m + σm
n∈Fl s∈S

2169-3536 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2606404, IEEE Access

TABLE I
KEY VARIABLES USED IN THIS PAPER . As we know, to solve (8), any direct search method would
involve an exhaustive search of all possible channel allocation
Variables Explanation a for all SUs. It is apparent that such an approach incurs an
M MU set
M Number of MUs extremely high complexity due to the huge action space of a.
N SBS set As mentioned before, there are two kinds of interference
N Number of SBSs relationships and a competition relationship, i.e., the cross-
L Dedicated channel set tiered interference relationship between SUs and MUs which
L Number of dedicated channels
S SU set is related to the second constraint in (8) and the co-tiered in-
S Number of SUs terference relationships among SUs belong to different SBSs,
Sn The SU set associated with SBS n and a competition relationship among SUs associated with
πn Number of SUs associated with SBS n
xni The i’th SU in small-cell n
same SBSs which is related to the first constraint. Due to the
bm Allocated channel set of MU m complex interference relationship and constraints, it is hard for
as , axni Allocated channel set of SU s and SU xni the macro-cell to allocate channels to SUs directly.
b Channel allocation vector of MUs On one hand, compared with the user terminal equipments,
a Channel allocation vector of SUs
an Channel allocation vector of SUs in small-cell n SBSs have strong abilities of communication and computing.
Cn Occupied channel set of SBS n In many situations, SBSs are deployed by communication
p0 , pn Transmission powers of MBS and SBS n operators. Thus, to meet the cross-tiered interference constraint
l
g0,m l
, gn,m Channel gain from MBS, SBS n to MU m in l and achieve a high global QoE utility, SBSs have a strong
Fl SBSs’ coalition who have occupied channel l
σ2 Noise power
motivation to utilize the channel resource cooperatively and
R Achieved throughput altruistically. On the other hand, SUs are rational and naturally
γ SNIR selfish with limited communication and computing abilities, so
B0 Bandwidth of each channel they focus on the available channels in their associated SBSs
u QoE utility function
Vm QoE threshold of MU m and selfishly compete for the channels to optimize their own
Il Interference threshold of MUs in channel l QoE satisfaction.
U Global QoE utility of SUs To simplify and resolve the complicated multichannel allo-
A Available channel allocation space cation problem, we propose a joint matching-coalitional game
ψl The MU who has occupied channel l
φs The associated SBS of SU s theoretical scheme and divide the problem into two subprob-
C SBS channel allocation strategy lems, i.e., Q1: intra-cell channel allocation for SUs, and Q2:
Un Sum of satisfaction of all SUs in small-cell n inter-cell channel allocation for SBSs. We apply an iterative
ϵxni Matching utility of SU xni
approach which deals with intra-cell channel allocation for
εl Matching utility of channel l
Un Coalitional utility of SBS n SUs (Q1) and inter-cell channel allocation for SBSs (Q2)
separately and iteratively. The channel allocation vector can
be rewritten as a = (a1 , a2 , . . . , aN ). And for the inter-cell
Thus, we can formulate the optimal resource allocation channel allocation, the SBS channel allocation strategy can
problem in the small-cell network as the following optimiza- be denoted as C = (C1 , C2 , . . . , CN ). Since the number of
tion problem channel allocation objects are much less in subproblems Q1
and Q2 than the original problem (8), the proposed solution
max U (a),
a∈A is more tractable.
as ∩ as′ = ø, ∀s, s′ ∈ Sn , n ∈ N ;
s.t. ∑ (8) Game theory is an effective approach towards solving
l
pn gn,ψl
< I l , ∀l ∈ L, the above resource allocation problem involving multiple
n∈Fl
interactive objects. Thus, we formulate the intra-cell channel
where A is the available channel allocation space, ψl is the allocation as a many-to-one selfish matching game in which
MU who has occupied channel l. the players are SUs and channel agents, and formulate the
The above problem is obviously a NP-hard non-linear inter-cell channel allocation subproblem as a cooperative and
programming problem [24]. In the formulated problem, it is altruistical coalitional game in which the players are SBSs
noted that the optimization utility function (7) has no obvious separately. The proposed joint matching-coalitional game not
increasing and concave properties with a even with the linear only focuses on optimizing a global function at the SBS level
constraint. Moreover, the cross-tiered interference constraint but also considers SUs’ selfish decisions and local interactions
makes this problem more difficult. Therefore, it is too hard to at the SU level. The specific schematic of the joint matching-
derive the optimal solution by gradient descent algorithm or coalitional game theoretical scheme is shown in Fig.2. In the
some other convex optimization algorithms. following, we will analyze the two subproblems separately.
A summary of key notations is presented in Table I.
A. Many-to-one matching game for intra-cell channel alloca-
III. A J OINT M ATCHING - COALITIONAL G AME tion
T HEORETICAL S CHEME FOR R ESOURCE A LLOCATION
In this subsection, we propose an intra-cell channel allo-
In this section, we resort to a joint matching-coalitional cation matching game to solve subproblem Q1 [25]1 . Given
game theoretical scheme to solve such a complicated problem
of QoE based multichannel resource allocation. 1 Part of this work was accepted by IEEE/CIC ICCC2016.

2169-3536 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2606404, IEEE Access

Cross-tiered interference the transmitters and available resources that are individually
constraint
conducted by such preference lists.
Definition 1: A matching µ is defined as a function from the
set Cn ∪Sn into the set of Cn ∪Sn such that (l ∈ Cn , s ∈ Sn ): 1)
Channel coalition µ(l) ∈ Sn and |µ(l)| = 1 for each channel, 2) µ(s) ∈ Cn ∪{ø}
and |µ(s)| ∈ {0, 1, 2, . . . , |Cn |} for each SU and 3) l ∈ µ(s)
Cooperative and altruistical if and only if µ(l) = s.
coalitional game Therefor, the tuple {Cn , Sn , ≻C , ≻S } determines the small-

x
x
x
cell channel allocation problem with ≻C = {≻l }l∈Cn being
Iteration
the preference set of the channels, ≻S = {≻s }s∈Sn being the
preference set of the SUs.
Many-to-one selfish To fully describe the matching µ, we define the preferences
matching games by two sides of the game. From the SUs’ side, each SU xni
(xni = s ∈ Sn ) can occupies multichannel and seeks to
Channel agent maximize its own, individual satisfaction function. Therefore,
xni only focuses on and applies for the channels which are
Fig. 2. The specific schematic of the joint matching-coalitional game not occupied by itself, i.e., Cn \ {axni }. It is assumed that
theoretical scheme.
each SU competes for only one of the best channel for its
satisfaction in Cn \ {axni } every time until the game reaches
a feasible macro-cell channel allocation C(t), we attempt to stability. Assuming that the current channel allocation vector
find an efficient channel assignment for SUs in each small- is an = (axn1 , axn2 , . . . , axnπn ), the satisfaction of SU xni
cell at iteration t. Specifically, for small-cell n the allocated can be rewritten by uni (axni | C−n ) = uxni (Rxni ), where
channel set is Cn (t). When SU xni (xni ∈ Sn ) transmits in C−n = (C1 , . . . , Cn−1 , Cn+1 , . . . , CN ).
channel l ∈ Cn (t), i.e., l ∈ axni , then the SINR of SU xni in Thus, the utility of SU xni in small-cell n for a given
channel l can be expressed as channel l ∈ Cn \ {axni } ∪ {ø} is given by
l
pn gn,x ϵxni (l, axni | C−n ) = uni (axni + l | C−n )
(11)
γxni ,l = l
∑ ni . (9) −uni (axni | C−n ),
p0 g0,x ni
+ pn′ gnl ′ ,xni + σ 2
n′ :l∈Cn′ (t),n′ ̸=n where uni (axni + l | C−n ) is the satisfaction of xni when
Thus, based on (6) the ∑ aggregate throughput of SU xni adds channel l to transmit. In particular, when l = ø, there
at iteration t is Rxni (t) = l∈ax B0 log2 (1 + γxni ,l ). For have ϵxni (ø, axni | C−n ) = 0.
ni
each small-cell n, it tries to maximize the sum of satisfaction For SU xni , a preference relationship ≻xni is defined that,
of all SUs in small-cell n, i.e., for any two channels l, l′ ∈ Cn \ {axni } ∪ {ø}, l ̸= l′ , there has
l ≻xni l′ if and only if ϵxni (l, axni | C−n ) > ϵxni (l′ , axni |

πn
Un (an | C(t)) = uxni (Rxni (t)). (10) C−n ).
i=1 From the channels’ side, each channel seeks to maximize
Because the macro-cell channel allocation C(t) is fixed at its own, individual contribution to the SU satisfaction. Specifi-
iteration t and small-cell channel allocation an (t) of n does cally, from the channel l’s point of view, there are two kinds of
not affect other SBSs n′ ∈ N \n, all SBSs can optimize their SUs in small-cell n, i.e., i) the SU who have occupied channel
channel allocation simultaneously. l in the current and ii) the other else SUs in small-cell n.
|C (t)| Then, the utility of the channel l in small-cell n for a given
There are πn n small-cell channel allocation strategies
for small-cell n. When πn or |Cn (t)| is large, the action space SU xni is given by
of an is huge, and direct search methods can not handle this ε{l (xni | C−n )
subproblem effectively. In the following, we resort a many-to- uni (axni + l | C−n ) − uni (axni | C−n ), l ̸∈ axni
=
one matching game to solve subproblem Q2. uni (axni | C−n ) − uni (axni − l | C−n ), l ∈ axni ,
The small-cell channel allocation approach using matching (12)
game involves multiple decision-making agents, i.e., the avail- where where uni (axni − l | C−n ) is the satisfaction of xni
able channel resources (Cn (t)) and the SUs in small-cell n when abandons channel l from axni .
(Sn ). The solutions (i.e., matching between channel resources Thus, for channel l, a preference relationship ≻l is defined
and SUs) are produced by individual actions of the agents. that, for any two SUs xni , xni′ ∈ Sn , i ̸= i′ , there has xni ≻l
In the following proposed many-to-one matching game, the xni′ if and only if εl (xni | C−n ) > εl (xni′ | C−n ).
available channel resources (Cn (t)) will be assigned to a set It is shown that the preference relationships of SUs and
of SUs (Sn ), where each channel will be assigned to at most channels are strongly affected by the current formation of other
one SU. In the following section, we replace Cn (t) with Cn SU-channel matching due to externalities of the utility func-
for convenience, as well as other notations in small-cell n. tions. Due to externalities, SUs and channels may continuously
The actions (matching requests, confirmation or rejection) change their preference orders. Thus, the traditional deferred
are determined by the given preference profiles. The match- acceptance (DA) algorithm based on preference orders without
ing outcome yields mutually beneficial assignments between externalities are unsuitable as the ranking of the preference

2169-3536 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2606404, IEEE Access

Algorithm 1: Intra-cell channel allocation algorithm using SBSs are modeled as the players. To investigate this subprob-
stable matching lem, we denote the coalitions as F = (F1 , F2 , . . . , FL ), where
Fl ⊂ N . In other words, all the SBSs who occupy channel
Initialization: At current time t, calculate C based on channel l constitute the coalition Fl and each SBS n can participate
selection profile a. in multiple coalitions simultaneously. Consequently, all SBSs,
Matching algorithm: who interfere with each other in channel l, are in the same
Step 1: Each SU xni calculates the utility ϵxni (l, axni | C−n ), coalition Fl , thus the co-tier interference can be efficiently
l ∈ Cn \ {axni } ∪ {ø} and updates its preference ≻xni . Each mitigated inside the coalition Fl .
SU xni sends a proposal to its most favored channel. Given a channel allocation vector a = (a1 , a2 , . . . , aN )
Step 2: Each channel l (l ∈ Cn ) calculates the utility εl (xni | and the corresponding macro-cell channel allocation strategy
C−n ) for all the SUs who have sent proposals to channel l and C = (C1 , C2 , . . . , CN ), then the coalitions are determined, i.e.,
the SU who has occupied channel l and updates its preference F (a) = (F1 (a), F2 (a), . . . , FL (a)) in which Fl (a) = {n :
≻l . l ∈ Cn }.
Step 3: Randomly select a channel l from Cn , and withdraw Formally, we define a coalitional game among the SBSs as
channel l from the SU (xni′ ) who has occupied l before, {N , F , U}, where each SBS n can i) join in a coalition Fk
i.e., axni′ = axni′ − l. Then, redistribute l to the first (n ̸∈ Fk ), ii) depart from a coalition Fl (n ∈ Fl ) or iii) replace
preferential SU (xni∗ ) among the proposal SUs and xni′ , i.e., another SBS n′ in a coalition Fh (n′ ∈ Fh , n ̸∈ Fh ).
axni∗ = axni∗ + l. Update the channel allocation vector an . When SBS n tries to join in a coalition Fk (n ̸∈ Fk ), each
Output: Return to Step 1 and repeat until ̸ ∃[l, s], s.t. l ≻s ø SU in small-cell n calculates the increment of QoE utility
and s ≻l µ(l). Then stable matching µ. (ϵxni (l, an | C−n )) after adds channel l to transmit based
on (11). If SBS n has joined in the coalition Fk , then SBS n
allocates channel k to the SU with the greatest QoE increment,
i.e., xni∗ = argxni max ϵxni (k, an | C−n ) and axni∗ ,k =
varies as the matching forms. We propose a small-cell channel axni∗ ∪ {k}. Then, the individual utility of SBS n to join
allocation algorithm that performs a matching between the in coalition Fk (n ̸∈ Fk (a)) is defined as follows
channels and SUs while taking into account the individual  ∑
satisfactions. 
 {Uj (a′ ) − Uj (a)},

 j∈Fk (a)∪{n}
Definition 2: A matching µ is competitive stable, if and only ∑
Un (+k, a) = pn gn,ψk + pj gj,ψk < I k
if no pair of {[l, s] | l ∈ Cn , s ∈ Sn } blocks the matching. That 



j∈Fk (a)
is, ̸ ∃[l, s], s.t. l ≻s ø and s ≻l µ(l). 0, otherwise,
(13)
Theorem 1. The proposed intra-cell channel allocation algo-
where a′ = a \ {axni∗ } ∪ {axni∗ ,k }. The defined utility
rithm is guaranteed to converge to a local stable matching.
function reflects the marginal contribution of ∑ SBS n joining
Proof: Due to the limited numbers of SUs and occupied coalition Fk . It is clear that when pn gn,ψk + pj gj,ψk <
channels, the sum of satisfaction of all SUs in small-cell n is j∈Fk (a)

limited. Moreover, the sum of satisfaction (10) is nondecreas- I k , the individual


∑ utility can be rewritten as ϵxni∗ (k, axni∗ |
ing with Step 3, i.e., εl (xni∗ | C−n ) − εl (xni′ | C−n ) ≥ 0. C−n ) + {Uj (a′ ) − Uj (a)}. Because each SBS can
j∈Fl (a)
Thus, after finite repeats, the proposed algorithm will finally transmit in multichannel and there must be ϵxni∗ (k, axni∗ |
converge to a global or local maximum point of Un with stable C−n ) ≥ 0, ∀n ∈ N , k ∈ L \ {Cn }, thus each SBS would
matching and no further improvement can be achieved in Step like to join coalitions as many as possible to increase its
3, i.e., xni∗ = xni′ . own utility. However, due ∑to the interference caused by each
SBS’s action, there has {Uj (a′ ) − Uj (a)} ≤ 0. Thus,
j∈Fl (a)
B. Coalitional game with overlapping coalitions for inter-cell each SBS should takes the utility loss of other members of
channel allocation the coalition into account when deicides whether to join the
The coalitional game, in which the players form coalitions coalition.
to improve the system performance, has recently been studied When SBS n departs from a coalition Fl (n ∈ Fl ), SU
in the area of wireless communications networks. Unlike xni′ , who had occupied channel l, has to give up to transmit
capacity limited and selfish SUs, SBSs have strong incentives in channel l, i.e., axni′ ,−l = axni′ \ {l}. Then, the individual
to cooperation with strong abilities of communication and utility of SBS n to depart from coalition Fl (n ∈ Fl (a)) is
computing. Since each small-cell could share the L dedicated defined as follows
channels simultaneously, we formulate the macro-cell channel ∑
Un (−l, a) = {Uj (a′′ ) − Uj (a)}, (14)
allocation problem as a coalitional game with overlapping
j∈Fl (a)
coalitions.
′′
In particular, depending on the SINR feedbacks, the SBSs where a = a \ {axni′ } ∪ {axni′ ,−l }. The defined utility
can decide to form coalitions so as to mitigate the co-tiered function reflects the marginal contribution of SBS n joining
interference and manage the cross-tiered interference to MUs coalition Fl . For every SBS n, we formally define the follow-
in cooperative manners. In the proposed coalitional game, the ing join and departure rule:

2169-3536 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2606404, IEEE Access

Definition 3: Join and Departure Rule Consider a channel Algorithm 2: Joint channel allocation algorithm for the
allocation vector a and the corresponding macro-cell channel matching-coalitional game
allocation strategy C = (C1 , C2 , . . . , CN ). 1) A SBS n has an
incentive to join coalition Fk (n ̸∈ Fk (a)) if Un (+k, a) > 0. Step 1: The initial channel allocation profile is a[0], where
2) A SBS n has an incentive to depart from coalition Fl (n ∈ as [0] = ø, Cn = ø ∀s ∈ Sn , n ∈ N .
Fl (a)) if Un (−l, a) ≥ 0. Step 2: At iteration t, the optimizing action is chosen accord-
As we know that the received interference of a certain MU ing the probabilities PQ1 = ζ and PQ2 = 1 − ζ, 0 < ζ < 1.
in channel h is limited, thus the size of each coalition is Specifically, randomly select a number x from [0,1], if x < ζ,
limited. When SBS n wants to ∑join coalition Fhh (n ̸∈ Fh (a)) then go to Step 3, otherwise, go to Step 4.
and there has pn gn,ψh + pj gj,ψh > I , SBS n can Step 3: In each small-cell n, redistribute the occupied intra-
j∈Fh (a) cell channels Cn using the matching algorithm in Algorithm
try to replace with a SBS n′ (n ∈ Fh (a)) to achieve a higher 1. Each SBS n reports the SUs’ channel selection profile an ,
system utility. Then, we define the following replacement rule: then update the channel allocation profile a[t+1] synthetically.
Definition 4: Replacement Rule Consider a channel allo- Return to Step 2 and repeat.
cation vector a , for coalition Fh , a SBS n (n∑̸∈ Fh (a)) can Step 4: Randomly select a small-cell n and a channel l. i)
replace SBS n′ (n′ ∈ Fh (a)) if i) pn gn,ψh + pj gj,ψh > ∑
If n ̸∈ Fl [t] and pn gn,ψl + pj gj,ψl < I l , SBS n tries
∑ j∈Fh (a) j∈Fl [t]
I h , ii) pn gn,ψh + pj gj,ψh < I h , iii) Un (+h, ã) > to join in coalition Fl based on the join and departure rule.
j∈Fh (a)\{n′ }
ii) If n ∈ Fl [t], SBS n tries to depart from coalition Fl
Un′ (+h, ã), where ã is the allocation vector after SBS n′
based on the∑join and departure rule. iii) If n ̸∈ Fl [t] and
departs from coalition Fh , i.e., ã = a \ {axn′ i′ } ∪ {axn′ i′ ,−l }.
pn gn,ψl + pj gj,ψl > I l , find another SU n′ (n′ ∈ Fl [t]),
Remark 1: If a SBS n unilaterally takes an action (join in or j∈Fl [t]
depart from a coalition, or replace another SBS in a coalition) SBS n tries to replace n′ in Fl based on the replacement rule.
based on the join and departure rule and the replacement rule, Update the channel allocation vector a[t + 1]. Return to Step
the global QoE utility of SUs increases. It is clear that the 2 and repeat.
∑ QoE utility of SUs can be also expressed as U (a) =
global
Un . i) When each SBS n joins in a coalition Fk (n ̸∈ Fk ),
n∈N
the change of the global QoE utility of SUs is U (a′ )−U (a) = in Step 3 and Step 4 in proposed Algorithm 2. Theorem 2 is
Un (+k, a) > 0. ii) Similarly, when SBS n departs from a proved.
coalition Fl (n ∈ Fl ), the change of the network satisfaction
utility is U (a′′ ) − U (a) = Un (−l, a) ≥ 0. iii) When SBS n IV. S IMULATION R ESULT
replaces another SBS n′ in a coalition Fh (n′ ∈ Fh , n ̸∈ Fh ),
A. Simulation scenario
the change of the global QoE utility of SUs is U (ã\{axni∗ }∪
{axni∗ ,h }) − U (a) = Un (+h, ã) − Un′ (+h, ã) > 0. In the following simulation study, the radius of the macro-
To find the stable channel allocation profile, we propose a cell is 30m, the radius of small-cells is 10m. SBSs and MUs
joint channel allocation algorithm for the matching-coalitional are randomly located in the circular coverage of the MBS.
game. At each iteration t, we select a subproblem (Q1/Q2) There are 8 channels and the bandwidth of each available
to optimize according predetermined probabilities. From the channel is 180 kHz. The maximal transmission powers of
two games G1 and G2, it can be noted that U can strictly MBS and SBSs are P0max = 43dBm and Pnmax = 23dBm
increasing along such better reply actions of players. We repeat respectively. Then the transmission powers of MBS and SBSs
the two kinds of processes in all iterations until no further in each channel are set as p0 = P0max /L and pn = Pnmax /L.
improvement is made. The noise power density is −174dBm/Hz. To model the
propagation environment, we only consider path loss for
Theorem 2. The proposed proposed Algorithm 2 for the joint analysis simplicity for small-cell network as many of previous
matching-coalitional game is guaranteed to converge to a studies [11]. For path-loss P L = 38.46 + 20log(dns ), where
stable channel allocation profile of both the proposed many- dns is the distance between SBS n and SU s. For simplicity, it
to-one matching game and coalitional game jointly. is assumed that each MU has only occupied one of all available
channels, and the cross-tiered interference threshold of each
Proof: For intra-cell channel allocation in each small-
MU is randomly selected from zero to half of the received
cell n using matching algorithm in each Step 3, it is seen
MBS signal power. There are 1∼4 SUs associating with each
that the small-cell satisfaction Un is nondecreasing, and the
SBS. The required transmission rates of SUs are randomly
network satisfaction utility is increasing until achieves a global
distributed in [0, 2]Mbps, and the demand degree τ of each
stable matching. Moreover, it is seen that the updating SBS
SU is randomly select from [0, 20]. A simulation snapshot
n’s action in each Step 4 makes the global QoE utility U
with a MBS and 15 SBS is shown in Fig. 3.
increasing from the definitions of join and departure rule and
replacement Rule. Since the global QoE utility is up bounded,
the learning algorithm will finally converge to a global or B. Convergence analysis
local maximum point of U in finite steps, i.e., a stable channel We analyze the expected convergence behaviors of Algo-
allocation profile, and no further improvement can be achieved rithm 2. Here, we set 1) ζ = 0, 0.2, 0.4, 0.6 as constants and

2169-3536 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2606404, IEEE Access


1

0.9

The cumlative distribution function (CDF)


 0.8 ζ=0
ζ=0.2
0.7
 ζ=0.4
0.6 ζ=0.6
 ζ(t)
0.5

 0.4
MBS 0.3

MU 0.2
 SBS
0.1
SU
 0
         0 200 400 600 800 1000
Iterations
Fig. 3. A simulation snapshot with a MBS and 15 SBS.
Fig. 5. The convergence speed comparison.

16
30
14
Alogrithm 2
25 Smallest interference channel selection
The global QoE utility of SUs

12
Random selection
ζ=0
10 ζ=0.2
The global QoE utility
20
ζ=0.4
8 ζ=0.6
ζ(t) 15
6

4 10

2
5

0
0 100 200 300 400 500 600 700 800
Iterations 0
5 10 15 20
Number of SBSs

Fig. 4. The expected convergence behaviors of Algorithm 2 with ζ =


0, 0.2, 0.4, 0.6 and ζ(t). Fig. 6. The global QoE utilities when varying the number of SBSs.

2) ζ(t) = 1 − Tmaxt
where Tmax is a constant no less than the C. Global QoE utility analysis
maximum iterations (Here we set Tmax = 2000). As we know, In this subsection, we present the performance evaluation
different values of ζ mean different selection probabilities of results of achieved global QoE utility denoted as (7). For
intra-cell channel allocation and inter-cell channel allocation comparison, the achieved global QoE utilities are calculated by
in each iteration. The expected convergence behaviors of three different approaches: i) optimize the global QoE utility
Algorithm 2 are shown in Fig. 4. with the proposed Algorithm 2 (ζ = 0.1), ii) a randomly
It is noted from the figure that when the value of ζ decreases selected SBS allocates one of its associated SU the best
and ζ > 0, the convergence speed of Algorithm 2 significantly channel (the channel with the smallest received interference of
increases. The converged global QoE utility of Algorithm the SU) in each iteration, iii) each SU is randomly allocated
2 with ζ = 0 is smaller than all others. In other words, an available channel set at once with the constraints.
the performance of optimizing the global QoE utility with Firstly, we vary the number of SBSs in the heterogeneous
Algorithm 2 containing intra-cell channel allocation (ζ > 0) is wireless network, i.e., we set N = 5, 10, 15, 20 respectively
better than Algorithm 2 without intra-cell channel allocation and L = 8, max(πn ) = 4. The global QoE utilities when
(ζ = 0). varying the number of SBSs are shown in Fig. 6.
The companion results of the cumulative allocation function It is shown from Fig.6 that the utility performance of opti-
(CDF) of the iterations needed for converging are shown mizing the global QoE utility with the proposed Algorithm 2 is
in Fig. 5. Then, it can be concluded that the convergence the better than the smallest interference channel selection, and
performance of ζ(t) is the best. The reasons are that i) at the the utility performance of random selection is the worst. The
beginning of the iteration, intra-cell channel resource is scarce reason is that, the random selection only allocates channels
and ii) the globe utility improvement of inter-cell channel at once without any optimization process. In the smallest
allocation is much bigger than the improvement of intra-cell interference channel selection only one SBS can take action
channel allocation at most time before the convergence. in each iteration, while in Algorithm 2 SBSs can collaborate

2169-3536 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2606404, IEEE Access

16
20

14 18

12 16
The global QoE utility

14

The global QoE utility


10
12
8
10

6 8

Algorithm 2 6
4 Algorithm 2
Smallest interference channel selection
Random selection 4 Smallest interference channel selection
2 Random selection
2
0
6 7 8 9 10 0
3 4 5 6
Number of channels
The maximal SU number of each small cel

Fig. 7. The global QoE utilities when varying the number of channels. Fig. 8. The global QoE utilities when varying the maximal SU number.

with each other more fully. It is also noted from the figure game theoretical scheme to solve such a complicated multi-
that as the number of SBSs increases (the number of SUs also channel allocation problem. Based on the different interference
increase), the achieved utility of the proposed Algorithm 2 and competition relationships among users, we divided the
increases. complicated problem into two subproblems, i.e., Q1: intra-
Secondly, we vary the number of channels (for simplicity, cell channel allocation for SUs and Q2: inter-cell channel
the number of MUs varies with the number of channels) in the allocation for SBSs. We formulated the intra-cell channel
heterogeneous wireless network, i.e., we set L = 6, 7, 8, 9, 10 allocation as a many-to-one selfish matching game in which
respectively and N = 10, max(πn ) = 4. The global QoE SUs selfishly compete for the channels which have been allo-
utilities when varying the number of channels are shown in cated to SBSs, and formulated the inter-cell channel allocation
Fig. 7. subproblem as an altruistical coalitional game in which SBSs
It is noted from Fig. 7 that as the number of channels utilize the available channels cooperatively and altruistically
increases, the achieved utility also increases. It is also is under the cross-tiered interference constraint separately. Then
shown from the figure that the utility gap between Algorithm 2 the complicated problem could be solved based on the two
and the smallest interference channel selection is smaller and proposed games iteratively. We proposed a joint channel allo-
smaller as the number of channels increases. The reason is that, cation algorithm for the matching-coalitional game theoretical
as the channel resource increases, each SU has more available scheme. We proved that the proposed algorithm converges to
channels to be allocated, and the advantage of Algorithm 2 a stable channel allocation profile of both the proposed many-
is smaller. It is clear that if the channel resource is enough, to-one matching game and coalitional game. In the future, we
all SUs can be served with enough channel resource with will consider sociality-aware resource allocation problems in
Algorithm 2 and the smallest interference channel selection. 5G HCNs.
Thirdly, we vary the maximal SU number of each small-cell
in the heterogeneous wireless network, i.e., we set max(πn ) =
3, 4, 5, 6 respectively and N = 10, L = 8. In other words, R EFERENCES
there are 1∼max(πn ) SUs associating with each SBS. The [1] A. Osseiran, F. Boccardi, V. Braun, K. Kusume, , et al., “Scenarios
global QoE utilities when varying the number of channels are for 5G mobile and wireless communications: the vision of the METIS
shown in Fig. 8. project,” IEEE Communications Magazine, vol. 52, no. 5, pp. 26–35,
May 2014.
It is noted from Fig. 8 that as the maximal SU number [2] V. Chandrasekhar, J. Andrews, and A. Gatherer, “Femtocell networks:
increases, the achieved utility increases. The reason may be a survey,” IEEE Communications Magazine, vol. 46, no. 9, pp. 59–67,
that the number of SUs who have low channel resource Sep. 2008.
[3] Y. Sun, R. CJover, and X. Wang, “Uplink interference mitigation for
requirements also increase with the increase of the maximal OFDMA femtocell networks,” IEEE Transactions on Wireless Commu-
SU number. Thus, SBSs with the limited channel resource nications, vol. 11, no. 2, pp. 614–625, Feb. 2012.
can provide service to more SUs. Moreover, it can be found [4] C. X. Wang, F. Haider, X. Gao, et al.,“Cellular architecture and key
technologies for 5G wireless communication networks,” IEEE Commu-
from the figure that the utility gap between Algorithm 2 and nications Magazine, vol. 52, no. 2, pp. 122–130, Feb. 2014.
the smallest interference channel selection is more and more [5] Z. Du, Q. Wu, P. Yang, “Dynamic user demand driven online network
bigger as the maximal SU number increases. selection,” IEEE Communication Letter, vol. 18, no. 3, pp. 419–422,
2014.
[6] P. Kulkarni, W. H. Chin, and T. Farnham, “Radio resource management
V. C ONCLUSION considerations for LTE Femto cells,” ACM Sigcomm Computer Commu-
In this paper, we investigated the problem of QoE-based nication, vol. 40, no. 1, pp. 26–30, Jan. 2010.
[7] S. Y. Lien, C. C. Tseng, K. C. Chen, and C. W. Su, “Cognitive radio
multichannel allocation in 5G HCNs with cross-tiered inter- resource management for QoS guarantees in autonomous femtocell
ference constraint. We proposed a joint matching-coalitional networks,” inProc. IEEE Int. Conf. Commun., pp. 1–6, May 2010.

2169-3536 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2606404, IEEE Access

10

[8] H. C. Lee, D. C. Oh, and Y. H. Lee, “Mitigation of inter-femtocell


interference with adaptive fractional frequency reuse,” inProc. IEEE Int.
Conf. Commun., pp. 1–5, May 2010.
[9] Y. S. Liang, et al, “Resource allocation with interference avoidance in
OFDMA femtocell networks,” IEEE Transactions on Vehicular Technol-
ogy, vol. 61, no. 5, pp. 2243–2255, 2012.
[10] D. T. Ngo, S. Khakurel, T. Le-Ngoc, “Joint Subchannel Assignment and
Power Allocation for OFDMA Femtocell Networks,” IEEE Transactions
on Wireless Communications, vol. 13, no. 1, pp. 342–355, 2014.
[11] A. Abdelnasser, E. Hossain, and D. I. Kim, “Clustering and resource
allocation for dense femtocells in a two-tier cellular OFDMA network,”
IEEE Transactions on Wireless Communications, vol. 13, no. 3, pp.
1628–1641, 2014.
[12] S. He, Y. Huang, S. Jin, and L. Yang, “Coordinated beamforming
for energy efficienttransmission in multicell multiuser systems,” IEEE
Transactions on Communications, vol. 61, no. 12, pp. 4961-4971, 2013.
[13] C. Wang, et al, “Stackelberg game for spectrum reuse in the two-tier
LTE femtocell network,” in IEEE WCNC 2013, 2013.
[14] Y. Sun, et al, “Energy-aware joint user scheduling and power control
for two-tier femtocell networks: A hierarchical game approach,” IEEE
Systems Journal, 2016, doi:10.1109/JSYST.2016.2580560.
[15] Y. Sun, J. Wang, et al, “Local altruistic coalition formation game for
spectrum sharing and interference management in hyper-dense cloud-
RANs,” IET Communications, 2016, doi:10.1049/iet-com.2016.0094.
[16] S. Guruacharya, D. Niyato, E. Hossain, and D. I. Kim, “Hierarchical
competition in femtocell-based cellular networks,” in GLOBECOM
2010, pp. 1–5, Dec. 2010.
[17] X. Kang, R. Zhang, and M. Motani, “Price-based resource allocation for
spectrum-sharing femtocell networks: A stackelberg game approach,”
IEEE Journal on Selected Areas in Communications, vol. 30, no. 3, pp.
538–549, Apr. 2012.
[18] Y. H. Xu, J. L. Wang, Q. H. Wu, et al, “Opportunistic spectrum access
in cognitive radio networks: Global optimization using local interaction
games,” IEEE Journal of Selected Topics in Signal Processing, vol. 6,
no. 2, pp. 180–194, 2012.
[19] Y. H. Xu, Q. H. Wu, L. Shen, et al, “Opportunistic spectrum access with
spatial reuse: Graphical game and uncoupled learning solutions,” IEEE
Transactions on Communications, vol. 12, no. 10, pp. 4814–4826, 2013.
[20] Y. Li, D. Jin, J. Yuan, Z. Han, “Coalitional games for resource allocation
in the device-to-device uplink underlaying cellular networks,” IEEE
Transactions on Communications, vol. 13, no. 7, pp. 3965–3977, 2014.
[21] Y. Gu, W. Saad, M. Bennis, M. Debbah, “Matching theory for future
wireless networks: fundamentals and applications,” IEEE Communica-
tions Magazine, vol. 53, no. 5, pp. 52–59, 2015.
[22] L. Lei, Z. Zhong, C. Lin, and X. Shen, Operator controlled deviceto-
device communications in LTE-advanced networks,” IEEE Transactions
on Communications, vol. 19, no. 3, pp. 96C104, Jun. 2012.
[23] L. Chen, B. Wang, X. Chen, X. Zhang, and D. Yang, “Utility-based
resource allocation for mixed traffic in wireless networks,” in Pro. IEEE
INFOCOM WKSHPS, pp. 91-96, 2011
[24] T. H. Cormen, C. E. Leiserson, R. L. Rivest, and C. Stein,Introduction
to Algorithms, 2nd ed. Cambridge, MA, USA: MIT Press, 2001.
[25] D. C. Wu, Y. G. Tian, Y. B. Wang, and C. M. Wu, “QoE-based Dis-
tributed Multichannel Allocation in Small Cell Networks: A Potential-
matching Game Solution,” in Pro. IEEE/CIC ICCC’16, to appear.

2169-3536 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.